AgentPay

The checkout layer for the agent economy.

AgentPay is the AI-to-AI payment system for OpenClaw: agents can discover paid tools, complete USDC payments, and continue work—without human intervention.

Your specialist agent did the task. Another agent pays instantly. Work continues. Everybody wins. 🤝

Why AgentPay (and why now)

AI agents are moving from demos to real operations. The missing piece is simple:

How does one agent pay another agent for useful work over HTTP?

AgentPay answers that with a clean, developer-friendly flow built for OpenClaw-style systems.

• Monetize agent capabilities with per-endpoint pricing.
• Turn tools into revenue-generating services instead of internal-only utilities.
• Let agents transact autonomously (no manual invoices, no human-in-the-loop payment clicks).
• Keep orchestration smooth with automatic pay-and-retry.
• Get receipts and policy checks for auditability and control.

The pitch in one line

Stripe-like UX for agent-to-agent USDC payments over HTTP.

How it works

1. A seller agent protects an endpoint with a price.
2. A buyer agent calls it like normal.
3. Seller returns HTTP 402 + payment requirements.
4. Buyer agent pays and retries automatically.
5. Seller verifies, settles, and returns the result + receipt metadata.

Protocol headers:

• PAYMENT-REQUIRED
• PAYMENT-SIGNATURE
• PAYMENT-RESPONSE

Seller example (FastAPI)

from fastapi import FastAPI
from agentpay_seller import for_fastapi

app = FastAPI()
agentpay = for_fastapi(
    app,
    merchant_id="mrc_demo",
    api_key="ap_live_demo",
    base_url="http://localhost:8000",
    recipient="0xseller",
)

@app.post("/run-task")
@agentpay.protect(price="0.05")
async def run_task(payload: dict):
    return {"ok": True, "task": payload.get("task")}

agentpay.mount()

Buyer example

from agentpay_buyer import AgentPayClient

client = AgentPayClient.from_settings(settings)
response = await client.post("http://localhost:8001/run-task", json={"task": "summarize"})

print(response.json())
print(response.agentpay_receipt)

OpenClaw integration

AgentPay is purpose-built for OpenClaw workflows where agents call tools/services over HTTP.

Great fits:

• paid browser automations
• paid search/retrieval endpoints
• paid specialist sub-agent calls
• paid data prep/indexing/conversion jobs

A thin adapter is included at packages/integrations/openclaw/.

Quickstart

Install

pip install -e .

Run migrations

PYTHONPATH=packages/core alembic upgrade head

Seed demo data

PYTHONPATH=packages/core:apps/api python -m seed_demo

Start API

PYTHONPATH=packages/core:apps/api uvicorn main:app --reload --port 8000

Start seller demo

AGENTPAY_API_BASE_URL=http://localhost:8000 PYTHONPATH=packages/core:packages/seller_sdk:apps/seller_demo uvicorn main:app --reload --port 8001

Run tests

PYTHONPATH=packages/core:packages/buyer_sdk:packages/seller_sdk:packages/integrations/openclaw:apps/api pytest -q

Demo

Run examples/openclaw_paid_tool_demo.py for a minimal end-to-end flow: buyer agent calls paid tool → gets 402 → pays in USDC → retries → receives result + receipt.

License

See LICENSE.